基于极端梯度提升算法的地震同相轴自动识别  

Automatic detection of seismic event based on eXtreme gradient boosting

作　　者：黄建平[1] 张若枫 高睿语 李亚林[3] 段文胜[3] 陈飞旭 郭廷超[4] 潘成磊[4] HUANG Jianping;ZHANG Ruofeng;GAO Ruiyu;LI Yalin;DUAN Wensheng;CHEN Feixu;GUO Tingchao;PAN Chenglei(School of Geosciences in China University of Petroleum(East China),Qingdao 266580,China;SINOPEC Geophysical Research Institute,Nanjing 211100,China;Tarim Oilfield Branch,CNPC,Korla 841000,China;Geophysical Prospecting Research Institute of Jiangsu Oilfield Company,SINOPEC,Nanjing 210046,China)

机构地区：[1]中国石油大学(华东)地球科学与技术学院,山东青岛266580 [2]中国石化石油物探技术研究院,江苏南京211100 [3]中国石油塔里木油田公司,新疆库尔勒841000 [4]中国石化江苏油田分公司物探研究院,江苏南京210046

出　　处：《中国石油大学学报（自然科学版）》2024年第3期44-56,共13页Journal of China University of Petroleum(Edition of Natural Science)

基　　金：国家重点研发计划项目(2019YFC0605503);国家自然科学基金优秀青年科学基金项目(41922028);国家自然科学基金创新研究群体基金项目(41821002);山东省科研机构运费等专项(2021QNLM020001-5)。

摘　　要：在常规地震同相轴识别方法基础上,通过引入极端梯度提升算法(XGBoost)智能化策略,并结合地震数据相邻道相似性特征,发展一种基于极端梯度提升算法的地震同相轴自动识别技术方法。在编程实现方法的基础上,通过简单层状模型和复杂Marmousi模型模拟的记录进行测试,验证方法的正确性。对含噪音数据和实际资料中的同相轴进行识别测试,同时进行单道对比定量分析以及不同信噪比情况下算法预测结果精度对比。结果表明:新方法对含噪数据和实际资料均具有较好的适应性;在低信噪比(-6.98 dB)情况下,同相轴的查准率仍可超过90%。This paper presents an automatic seismic event detection method based on eXtreme gradient boosting(XGBoost) by integrating intelligent strategies and leveraging the similarity characteristics of adjacent traces in seismic data.The proposed method is validated through programming and testing on both simple layered and complex Marmousi models.The detection tests conducted on noisy data and the real data demonstrate the method's robustness and adaptability,even in low signal-to-noise ratio(SNR)conditions(-6.98 dB),achieving a seismic event detection precision of 90%.Additionally,single channel contrast quantitative analysis and comparison of algorithm prediction accuracy under various SNR conditions further confirm the method's feasibility and applicability.

关 键 词：同相轴拾取 机器学习 特征拾取 极端梯度提升算法 

分 类 号：P631.4[天文地球—地质矿产勘探]